segfault doing 'bzr log' on a stacked branch
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| Bazaar |
Fix Released
|
Critical
|
John A Meinel | ||
Bug Description
time bzr log lp:~jameinel/bzr/2.0b1-stable-groupcompress-order -n0 -r -20..-1
has a segfault.
I believe it is a couple compounding factors:
1) the new "VF.get_
2) The new KnownGraph.
File "C:\Users\
parent_info = revnos[parents[0]]
Which is strange, because the code *does* handle ghosts. So I don't know how we got a parent that wasn't numbered before its child, that wasn't a ghost.
I'm currently investigating, but this is a fairly serious regression for the 2.0-rc1 release.
Related branches
- Vincent Ladeuil: Approve
- Diff: 160 lines
- bzr-core: Pending requested
- Diff: None lines
- Vincent Ladeuil: Approve
- Diff: 105 lines
| John A Meinel (jameinel) wrote | #1 |
| Robert Collins (lifeless) wrote | #2 |
| Changed in bzr: | |
| status: | In Progress → Fix Released |
So I tracked the segfault down into code that is passing a node with parents=None to _push_node which isn't supposed to be possible. (The code in _push_node assumes that it can check the parents list as a tuple, etc.)
I've got a quick way to reproduce this, so that part of this bug should be fixed soon.
I don't have a great idea (yet) as to how to handle the issues with stacking and find_ancestry. I think the interface gives back missing keys, so it might not be too bad.
(How to handle it with bzr+ssh repos that don't have stacking set up... I really don't know)